Automatic Speech Onset and Offset Detection Using Short-Time Energy and Zero-Crossing Rate of Speech Signals

In MATLAB code implementation, we can utilize the short-time energy and short-time zero-crossing rate of speech signals to achieve automatic detection of speech onset and offset points. By analyzing variations in both energy and zero-crossing rate characteristics of the speech signal, we can accurately determine the starting and ending positions of speech segments. This method employs frame-based processing where the audio signal is divided into short overlapping frames, typically 20-30ms in duration. For each frame, the short-time energy is calculated as the sum of squared sample values, while the zero-crossing rate counts the number of sign changes within the frame. The algorithm then applies adaptive thresholding to these temporal features to identify speech boundaries. This approach demonstrates excellent performance in detecting speech endpoints and can be effectively applied to practical speech processing tasks such as endpoint detection in speech recognition systems, voice activity detection, and audio segmentation. Key MATLAB functions involved include buffer() for frame segmentation, mean() for energy calculation, and customized thresholding logic for decision making.